Investigating fine-scale spatio-temporal predator–prey patterns in dynamic marine ecosystems: a functional data analysis approach



1. Spatial management of marine ecosystems requires detailed knowledge of spatio-temporal mechanisms linking physical and biological processes. Tidal currents, the main driver of ecosystem dynamics in temperate coastal ecosystems, influence predator foraging ecology by affecting prey distribution and ecology. The mechanistic links between tidal currents and how they influence predator–prey behaviour and interactions at a fine scale are poorly understood.

2. Studies of fine-scale changes in oceanography, prey and predator behaviour with tidal currents require repeated surveys of the same location over brief time-scales. Such data are highly temporally and spatially autocorrelated and require appropriate analytical tools.

3. We used functional data analysis (FDA), specifically functional principal component analysis (FPCA), to analyse repeated, fine-scale, survey data collected in the North Sea. FPCA was used to explore the relationship between the behaviour of an important North Sea prey species (sandeel Ammodytes spp.) and a vulnerable surface-foraging predator (black-legged kittiwake Rissa tridactyla) with fine-scale tidally driven changes in bio-physical characteristics (temperature stratification and maximum subsurface chlorophyll concentration).

4. The FPCA indicated that sandeels were aggregated close to the surface at maximum ebb (ME) currents. Surface-feeding kittiwakes were also found in highest numbers during ME in locations of both high subsurface chlorophyll concentration and shallow sandeel aggregations. We suggest that the combination of a well-stratified water column with the movement of tidal currents over uneven topography results in surface aggregations of sandeels which kittiwakes exploit.

5.Synthesis and applications. Functional Data Analysis provides a useful tool for examining spatio-temporal patterns in natural ecosystems. In combination with fine-scale repeated survey design, we identified the importance of tide in driving prey behaviour and hence predator foraging behaviour. This has implications both for critical marine habitat identification for Marine Protected Area selection and for fisheries stock assessments. We therefore recommend that tidal aspects should be taken into account when designing marine surveys in temperate coastal ecosystems both to ensure the best identification of critical marine habitat and to improve the accuracy of fish stock assessments.


Ecosystem management requires understanding of the functioning of marine ecosystems, from larger-scale dynamics down to species interactions (Botsford, Castilla & Peterson 1997; Barange 2005). In particular, the spatial management of the marine environment, whether through an ecosystem approach to fisheries management (EAFM), designation of Marine Protected Areas (MPAs) or offshore renewable developments, should be based on a detailed knowledge of the temporal and spatial mechanisms linking physical and biological processes (Botsford, Castilla & Peterson 1997; Agardy 2000). However, understanding interactions between dynamic physical environments, predators and prey is challenging especially in the marine environment where processes are hidden from view.

In high-latitude coastal ecosystems, the main driver of ecosystem dynamics is tidal current, influencing everything from primary productivity through zooplankton, fish and top predators (e.g. Hunt et al. 1998; Zamon 2002, 2003; Sharples et al. 2007). Studies have shown that the movement of tidal currents over topography acts to aggregate prey both vertically and horizontally (Zamon 2002; Genin 2004; Gόmez-Guitiérrez, Martínez-Gόmez & Robinson 2007), resulting in spatially and temporally predictable patchy prey distributions (Riley 1976). Such predictability facilitates predator–prey interactions and highlights critical areas for management (Piatt et al. 2006). Yet most studies are based on data collected along repeated transects with long time intervals rather than continuously (e.g. Hunt et al. 1998; Zamon 2003; Ladd et al. 2005). We have only identified a few other studies that repeatedly surveyed the same transects continuously over a complete diurnal/tidal cycle (Robinson, Gómez-Aguirre & Gómez-Gutiérrez 2007; Bertrand et al. 2008). By surveying the same area repeatedly over a tidal cycle, it is possible to identify fine-scale changes in bio-physical characteristics and prey behaviour with the diurnal/tidal cycle and to investigate how these changes impact predator foraging behaviour.

Information collected continuously over repeated transects, however, provides data that are temporally and spatially autocorrelated, requiring appropriate statistical analysis techniques. Functional data analysis (FDA), a relatively new statistical approach robust to autocorrelated data, provides tools for describing and modelling sets of functions (or curves) rather than vectors of single values (Ramsay & Silverman 2006). FDA has been used to examine functional data in many fields from economics (Ramsay & Ramsey 2002) to genetics (Illian et al. 2009). In our study, each repeat of the survey area was treated as a separate function in the FDA, allowing examination of patterns of change in bio-physical characteristics (subsurface chlorophyll concentration and temperature stratification), predator (black-legged kittiwakes Rissa tridactyla) and prey (sandeels Ammodytes spp.) behaviour in the Marr Bank region of the northwestern North Sea over the diurnal/tidal cycle.

Black-legged kittiwakes are particularly vulnerable to changes in the marine ecosystem, with declines in their breeding success linked to fisheries exploitation of their sandeel prey (Daunt et al. 2008) and to climate change (Wanless et al. 2007). This is believed to be due to their reliance on juvenile sandeels during the breeding season (Harris & Wanless 1997) and the limitations of surface-feeding constraints (Furness & Tasker 2000). Kittiwakes are therefore considered to be good indicators of environmental change in the North Sea ecosystem, being closely linked to the success of sandeels and hence their Calanus prey (Wanless et al. 2007).

The foraging success of kittiwakes is dependent on surface aggregations of sandeels, although we know very little about how these predator–prey interactions take place. The northwestern North Sea is tidally energetic and shallow, and the oceanographic climate of the study area around the Marr Bank region is mainly dominated by local tides and meteorology (Sharples et al. 2006). In these stratified waters during the late spring and early summer, a subsurface chlorophyll maximum (SCM) is commonly observed (Weston et al. 2005). A high SCM has been linked to foraging aggregations of many top predators in the North Sea, including kittiwakes (Scott et al. 2010). Chlorophyll provides a proxy for the feeding opportunity for higher predators, especially for those higher predators foraging on zooplanktivores such as sandeels (Steingrund & Gaard 2005). Thermal stratification has been linked to both kittiwake and sandeel abundance (van der Kooij, Scott & Mackinson 2008; Scott et al. 2010). The potential for seasonal stratification is defined by the tidal stratification measure h/U3 (where h is the water depth and U is the tidal current amplitude), where high values occur in areas that thermally stratify in summer. Marr Bank lies mostly within a region of high tidal stratification (log10(h/U3) > 3·55 m−2 s3) which has been associated with high values of subsurface chlorophyll (Scott et al. 2010). We know little about tidal influence on either sandeel schooling behaviour or kittiwake foraging behaviour, or how tides influence their interaction. Therefore, to understand how and when kittiwakes forage on sandeels, we need to understand how tides influence the bio-physical features (SCM and thermal stratification), sandeel schooling behaviour and kittiwake foraging behaviour.

Our objective was to use FDA to examine spatio-temporal patterns of bio-physical characteristics, sandeel and kittiwake behaviours at fine spatial (kilometres) and temporal (hours) scales. We used a circular sampling design to resample the same area repeatedly at fixed short intervals to investigate different aspects of tidal and daily rhythms. We tested the hypothesis that tidal speeds influenced the schooling behaviour of sandeels and the foraging behaviour of kittiwakes. We also examined whether kittiwakes selectively foraged on these tidally predictable surface aggregations of sandeels when combined with high subsurface chlorophyll concentrations. Our study represents a first attempt to use FDA to explore fine-scale spatio-temporal patterns in marine ecosystems.

Materials and methods

Data Collection

The survey was conducted in June 2003 from the 66-m research vessel R/V Pelagia as part of the EU IMPRESS project (Camphuysen 2005) and included the waters surrounding the Firth of Forth and the Isle of May within the North Sea, northeast Scotland (55·7–56·75ºN, 0·4–2·7ºW, Fig. 1). The Marr Bank region in which the study took place is well within the foraging range of breeding kittiwakes (Daunt et al. 2002). A large-scale survey covered eight transects in an east–west direction (Scott et al. 2010). Our study concentrated on three fine-scale circular surveys (mini-surveys, Fig. 1) carried out in three regions selected during the large-scale survey based on high numbers of foraging predators and contrasting bio-physical and topographic characteristics.

Figure 1.

 Locations of the three mini-surveys for the Pelagia cruise carried out from 12 to 15 June 2003 overlaid on bathymetry.

The circular surveys were designed so that the same locations were repeatedly sampled every sixth of the tidal cycle (2 h 5 min), with a survey path being a 10 by 2 km oval covered at a constant speed of eight knots. Each oval was surveyed 12 times within a 25-h period, covering two tidal cycles (c. one diurnal cycle). The timing of the circuits was designed to compare the fastest tidal speeds with the slower speeds during flooding and ebbing tides. The tidal cycle was divided into six states based on speed and direction of tidal current (Scott et al. 2005), such that maximum flood or ebb was considered 85% of the maximum tidal speed in a northerly or southerly direction (Table 1). Mini-surveys were run on consecutive days from 12 to 15 June 2003 during calm weather (<2 Beaufort scale) and spring tides.

Table 1.  Tidal states based on the speed and direction of the tidal current
Tidal state (abbreviation)Description
Maximum Ebb (ME)>85% maximum northerly tidal current
Decreasing Ebb (DE)Decreasing from <85 to 0% of northerly tidal current
Increasing Flood (IF)Increasing from 0 to <85% of southerly tidal current
Maximum Flood (MF)>85% maximum southerly tidal current
Decreasing Flood (DF)Decreasing from <85 to 0% of southerly tidal current
Increasing Ebb (IE)Increasing from 0 to <85% of northerly tidal current

During each survey, synoptic monitoring of the water column, fishes and top predators was carried out. Bio-physical oceanographic sampling was carried out using a ScanFish, an undulating oceanographic vehicle, providing continuous vertical and horizontal information on temperature, salinity, density and chlorophyll fluorescence at 1 Hz. Chlorophyll fluorescence was measured continuously throughout the water column using a Chelsea Instruments Aquatrack MKIII fluorometer mounted on the ScanFish. Fisheries acoustic data were collected using a SIMRAD EK500 scientific echosounder (Bodholt, Ness & Solli 1989), operating at 38, 120 and 200 kHz. The echosounder was configured to ping at each frequency simultaneously, every 1 s, with pulse duration of 1 ms for each frequency. Data were logged from the echosounder to a personal computer with Myriax’s Echolog software (Myriax Ltd, Australia, Hobart, Tasmania).

Foraging seabirds and mammals were recorded visually based on standard strip-transect counts (Tasker et al. 1984) using 5-min unit periods which at eight knots covers 1·24 km in distance (Camphuysen et al. 2004). Observations were made from the roof of the vessel’s bridge (∼15 m above sea level) during steaming, by two observers operating a 300-m-wide transect on one side and ahead of the ship. Standard recording methods were slightly modified to record behaviour, distinguishing feeding or foraging birds from non-feeding individuals (Camphuysen & Garthe 2004).

Oceanographic Data Processing

To compare the continuous water column characteristics measured by the ScanFish to the 5-min bin observations of the visible top predators, summaries of the physical and biological characteristics of the water column were created for each observational 5-min bin. A thermal stratification index (ΔT) was created using the difference between the mean temperatures above and below the pycnocline over each 5-min bin (Scott et al. 2010). Fluorescence was calibrated against chlorophyll concentration (Scott et al. 2010). Subsurface maximum chlorophyll values (CHLmax) were defined as the highest concentration of chlorophyll biomass within each 5-min bin of observations regardless of where they occurred vertically in the water column (Scott et al. 2010). Tidal stratification, log10(h/U3), was calculated using the actual measured depth during the survey and the tidal velocities from the POLPRED tidal prediction model (Proudman Oceanographic Laboratory, NERC, UK). The tidal velocities used are the mean monthly depth-mean tidal speeds for June 2003 and so represent average tidal speeds over two spring-neap cycles (Scott et al. 2010).

Fisheries Acoustics Data Processing

Echoview (Myriax Ltd.) was used to process the acoustic data and extract information on sandeel schools and density. Data were analysed between 10 and 0·5 m above the seabed, to exclude the near-field effect of the transducers and bottom pixels, respectively. A series of algorithms were applied to the 120 kHz echogram to compensate for the offset between the beams of each frequency, allowing the pixels from multiple frequencies to be matched in time and space (Korneliussen et al. 2008). Sandeels reflect more energy at 120 kHz than 38 kHz (Johnsen, Pedersen & Ona 2009), so all pixels for which the backscatter strength was greater at 38 kHz than at 120 kHz were excluded. These sandeel filtered data were used to create a synthetic echogram that contained only sandeel backscatter. The Echoview school detection algorithm (Barange 1994) was then used to automatically identify and select sandeel schools. School characteristics (perimeter and depth) were exported and averaged in 5-min bins to correspond to the time intervals of the top predator data. Data from the filtered echograms were also integrated over 5 min × 5 m depth bins to determine the Nautical Area Scattering Coefficient (NASC is proportional to abundance: MacLennan, Fernandes & Dalen 2002). These vertically stratified distribution data were used to extract the depth layer which contained the maximum sandeel density (depthmaxNASC) for each bin.

There are some limitations to using fisheries acoustics, most notably being the lack of data shallower than 10 m, and the difficulty of detecting fish on or buried in the sediment. The shallowest schools were assumed to correspond to the abundance of sandeels in water <10 m. Examining plots of NASC by depth shows increases in NASC values in shallower water with corresponding decreases in NASC in deeper waters as tidal speeds increase, suggesting that this is a valid assumption (results not shown). Sandeels buried in the sediment are not considered relevant because we were only interested in those available for surface-foraging seabirds.

The EK500 echosounder used to collect the sandeel data was not calibrated during the survey, but a sensitivity analysis showed there were no adverse effects on the acoustic variables used in this study (Appendix S1, Supporting information).

Functional Data Analysis

Functional Data Analysis provides tools for describing and modelling sets of curves or functions (Ramsay & Silverman 2002, 2006; Illian et al. 2005, 2009; Illian, Penttinen & Stoyan 2008). We used a Functional Principal Component Approach (FPCA) to identify the main patterns of variation. To illustrate, we used the number of sandeel schools per 5-min bin by repeat for MS1 as an example (Fig. 2). Because each ‘repeat’ surveyed the same area, the repeated sets of 5-min bins were matched in space; thus, 25 positions were surveyed spatially per repeat (Fig. 2b). The raw data on which the FDA was conducted are shown in Fig. 2a with the number of sandeel schools by position shown for each repeat. Each repeat is classed as a ‘function’ of the number of sandeel schools over the survey. The data used in the analysis were measured discretely xij and smoothed to generate functions xi(s), where i is the repeat number, and j is the 5-min bin number defining the position within the survey. Cubic b-splines were used for this purpose (Ramsay & Silverman 2006). The degree of smoothing was determined by penalising overfitting and carefully investigating the resulting functions. Too much smoothing levels out important features in the functions but too little causes spurious variation to be picked up by the methodology (Ramsay & Silverman 2006). FPCA was used to explore the primary mode of spatial variation in the data and how this changes temporally. This is similar to traditional PCA except that the data represent sets of continuous functions instead of vectors of discrete values. The weight vector in PCA βk now becomes a vector of functions with the values βk(s). FPCA determines weight functions βk(s) (or principal components (PCs) which are hence also functions) that best explain the original data xi(s), for example, PC1 is the weight function β1 for the number of sandeel schools over the survey area after the mean across all 12 repeats has been removed (Fig. 2c). In the spatial context, this weight function is best interpreted when overlaid with topography (Fig. 2e). In this example, the number of sandeel schools is higher over the shallower regions, with this function accounting for 78% of the variation. The tidal repeats that have high PC1 scores (repeats 1, 2 & 12, Fig. 2d) have higher numbers of sandeel schools over the shallow areas of the survey than on average, particularly on the western leg (positions 5–11).

Figure 2.

 MS1: (a) number of sandeel schools by 5-min position for each repeat; (b) MS1 circuit showing the position of each 5-min bin; (c) corresponding functional principal component analysis showing the first two PCs (solid line = PC1, dotted line = PC2); (d) corresponding PC scores plot, (1–12 = repeat number); (e) topography over a single circuit.

Functional principal component analysis (FPCA) was implemented in the add-on package ‘fca’ within the free software package ‘R’ (R Development Core Team 2011). FPCA was carried out for each survey on (i) CHLmax; (ii) ΔT; (iii) depthmaxNASC; the number of (iv) sandeel schools; and (v) feeding kittiwakes. FPCA plots for each variable were overlaid to evaluate the spatial and temporal interactions between bio-physical characteristics, predators (kittiwakes) and prey (sandeels) behaviour for each survey.

To examine large-scale hydrographic differences in CHLmax, ΔT and h/U3 between the survey areas, Kruskal–Wallis and Mann–Whitney U tests (where non-normal) or analysis of variance tests (where normal) were run in Minitab (version 12.23).


Survey Area and Bio-physical Characteristics

Of the three surveys, mini-survey 1 (MS1) had the highest CHLmax (< 0·001, H = 388·94, d.f. = 2, N = 848) and mini-survey 2 (MS2) the lowest stratification (< 0·001, H = 423·27, d.f. = 2, N = 848) and lowest tidal stratification (< 0·001, F = 61·03, d.f. = 2, N = 148) (Table 2). Mini-survey 3 (MS3) appeared to straddle two water masses with more stratified high CHLmax water moving into the survey area with the ebb tides (+PC1) and out with the flood tides (−PC1) as illustrated by the FPCA (Fig. 3, Table S1, Supporting information). The plot of FPCA scores clearly shows the spatial movement of water as it progresses through the tidal cycle (Figs 3b,d).

Table 2.  Summary of survey hydrographic variables
  1. Significant differences between MS1-MS2, MS2-3 and MS3-1 are shown with *< 0·05; **< 0·01; ***< 0·001.

Depth (m)Range47–6447–7163–75
Stratification (°C) Median3·7***2·9***3·7
Max chlorophyll (mg m−3)Median3·2***0·830·83***
Tidal stratification h/U3 (m−2 s3)Median3·583·56***3·62***
Number of sandeel schools 207685141
Number of foraging kittiwakes 1040389898
Figure 3.

 Functional principal component analysis of (a, b) ΔT, and (c, d) CHLmax for MS3; (a, c) first two PCs (solid line = PC1, dotted line = PC2), and (b, d) PC scores plot (1–12 = repeat number). Arrows show changes in PC scores progressing between repeats.

Temporal and Spatial Sandeel Behaviour

Sandeel schools were found in all three surveys, but in highest densities in MS2 (Table 2). FPCA of the number of sandeel schools showed where the main aggregations of sandeels were located in each of the survey areas (PC1, Fig. 4). In MS2, schools were found relatively uniformly throughout the survey area (PC1, Fig. 4), especially around the morning maximum flood when schools were most abundant (69% of all schools were found during repeats 8–11). FPCA for MS1 and MS3 showed sandeel schools clumped spatially (PC1 & PC2, Fig. 4).

Figure 4.

 Functional principal component analysis of number of sandeel schools for the three surveys: (d, e, f) first two PCs (solid line = PC1, dotted line = PC2), and (a, b, c) PC scores plots (1–12 = repeat number) for (a, d) MS1; (b, e) MS2; and (c, f) MS3. Corresponding fluorescence profiles for (g) MS1 repeat 12; (h) MS2 repeat 8; and (i) MS3 repeat 7.

Functional principal component analysis of the depth of maximum sandeel NASC (depthmaxNASC) showed some temporal patterns in all three surveys (Fig. 5; Figs S1 and S2, Supporting information). Maximum (ME) and decreasing ebb (DE) separated out on the PC plots, with a +PC2 in MS1; −PC1 in MS2; and −PC1 in MS3 (Fig. 5d; Figs S1d and S2d, Supporting information). Spatially, this showed that depthmaxNASC was shallow over the bank in MS1 and shallow over most of MS2 and MS3 during ME and DE (Fig. 5c; Figs S1c and S2c, Supporting information). In MS1, the depthmaxNASC also separated out by time of day: thus, at tidal speeds other than DE and ME, there were different spatial patterns of depthmaxNASC during daylight (+PC1) than at night (−PC1, Fig. 5d). This meant that during the day, depthmaxNASC was generally deeper than at night, except at DE and ME when it was always shallower on the bank regardless of time of day (Fig. 5c).

Figure 5.

 Functional principal component analysis of MS1 for (a, b) number of feeding kittiwakes, (c, d) depthmaxNASC, and (e, f) CHLmax where (a, c, e) show the first two PCs (solid line = PC1, dotted line = PC2), and (b, d, f) the corresponding PC scores plot (1–12 = repeat number); and (g) temperature profile for repeat 4.

Temporal and Spatial Kittiwake Foraging Behaviour

Feeding kittiwakes were found in all three surveys, but in highest densities in MS1 and MS3 (Table 2). Within these mini-surveys, 100% of kittiwakes were associated with Mixed-Species Feeding Associations (MSFAs) in MS2, but only 62–69% in MS1 and MS3, respectively. In these two surveys, the highest number of feeding kittiwakes was found at ME. FPCA of the number of feeding kittiwakes explained a very high amount of their variation in MS1 (PC1 explained 78%) and MS3 (PC1 explained 90%). In both of these surveys, the feeding kittiwakes were spatially clumped into large aggregations (Fig. 5a and Fig. S2a, Supporting information), especially in MS3 where there was a single aggregation of 353 feeding kittiwakes in one location at the southern end of the survey area. The high amount of variation explained by a single PC may be partly explained by the temporal aggregations of feeding kittiwakes – with the kittiwakes mainly clumped into feeding aggregations at ME.

MS2 showed very little temporal variation in numbers of feeding kittiwakes. Both PCs suggested some spatial aggregation of feeding kittiwakes into two main locations at the southwest end of the survey and to the north of the survey area (Fig. S1a).

Interactions Between Kittiwakes, Sandeels and Bio-physical Characteristics

Comparing the FPCA results for the number of kittiwakes, the depthmaxNASC and CHLmax, some patterns are revealed (Fig. 5 and Figs S1 and S2, Supporting information). In MS1 and MS3, maximum kittiwake foraging coincided spatially and temporally with shallow depthmaxNASC and high CHLmax. In MS1, the highest number of feeding kittiwakes was found during the afternoon ME (repeat 3). During this repeat, the FPCA for depthmaxNASC is defined by a +PC2 (Fig. 5d) showing a shallow depthmaxNASC where the kittiwakes are found (Fig. 5a,c). Also, the FPCA for CHLmax shows a +PC1 and +PC2 (Fig. 5f), matching spatially with a high CHLmax where the kittiwakes are feeding (Figs 5a,e). In MS3, the highest number of feeding kittiwakes was found during morning ME (repeat 7). During this repeat, the FPCA for depthmaxNASC has a −PC1 (Fig. S2d) showing a shallow depthmaxNASC where the kittiwakes are found (Figs 5a,c). In addition, the FPCA for CHLmax has a +PC1 (Fig. 5f), resulting in a spatial match between high CHLmax and the location of high foraging kittiwakes (Figs 5a,e). This change of behaviour by sandeels and kittiwakes with CHLmax is visible as the tide changes from increasing Ebb (IE) to ME (Fig. 6), sandeels becoming shallower and more clumped as the tide progresses from IE to ME, with a corresponding rise in CHLmax and the number of feeding kittiwakes.

Figure 6.

 MS3 during morning increasing (a, c, e) and maximum ebb tides (b, d, f) showing (a, b) number of feeding kittiwakes per 5-min bin; (c, d) corresponding EK500 echogram from 120 kHz showing sandeel schools at −70 dB; and (e, f) fluorescence profile for corresponding repeat showing current direction, and a box indicating the area shown in (a, b, c, d).

In contrast to the other mini-surveys, there were no clear spatial or temporal interactions between feeding kittiwakes, sandeels and bio-physical characteristics in MS2 (Fig. S1, Supporting information).


In our study, we used FDA to demonstrate the importance of tidal currents in the physical–biological coupling of trophic transfer to top predators. Physical–biological coupling is an important aspect of the marine ecosystem influencing everything from primary production (e.g. Weston et al. 2005) to trophic transfer to fish and top predators (Hunt et al. 1998; Genin 2004; Bertrand et al. 2008). With the current concerns over the impact of both fisheries and climate change on the functioning of marine ecosystems, understanding the mechanisms of bio-physical coupling has been identified as one of the key knowledge gaps within both MPA designation and fisheries management. (Botsford, Castilla & Peterson 1997; Agardy 2000; Barange 2005). In the UK, the kittiwake–sandeel ecosystem is considered a classic example of this issue (Barange 2005), with declines in kittiwake breeding success linked to both fisheries exploitation of their sandeel prey (Daunt et al. 2008) and to climate change (Wanless et al. 2007). Thus, a better understanding of the bio-physical mechanisms linking these species would help in an ecosystem approach to their conservation. Also, tidal ecosystems are of particular interest because the interaction between tides and topography results in spatially and temporally predictable aggregations of prey that predators can exploit (Decker & Hunt 1996; Piatt et al. 2006). More generally, this temporal and spatial predictability of prey resources can help in marine spatial planning activities such as MPA designation and placement of marine renewable devices. We believe our study is the first to simultaneously record bio-physical characteristics and predator–prey distributions and behaviours to examine bio-physical coupling within tidal ecosystems.

Using FDA, in particular FPCA, we identified changes in both sandeel schooling behaviour and kittiwake foraging behaviour with tidal currents. In all surveys, sandeels were shown to be shallow at maximum and DE regardless of time of day. Aggregation of sandeels close to the surface at ME suggests that their zooplankton prey is similarly aggregated close to the surface at this time. Tidal currents flowing over uneven topography such as banks have been widely reported to upwell zooplankton towards the surface, causing localised aggregations of shallow zooplankton (Zamon 2002; Genin 2004; Ladd et al. 2005). This results in higher feeding aggregations of fish close to the surface at maximum tidal currents (Zamon 2002). Thus, sandeels may be maximising their foraging opportunity by gathering close to the surface at ME when the food is most concentrated. ME therefore represents the best trade-off between maximising sandeel foraging opportunity and minimising their risk of predation, similar to other studies of fish (e.g. Regular et al. 2010). Shallower sandeels also result in a higher abundance of sandeels in the surface 10 m, unavailable to the fisheries acoustics. This has implications for sandeel stock assessment surveys, which use both acoustics and bottom grab samples to determine overall sandeel biomass (e.g. Greenstreet et al. 2006). If sandeels are neither in the sediment nor available to fisheries acoustics around ME, then the biomass of sandeels could be underestimated.

Based on the high numbers of foraging kittiwakes observed during ME in MS1 and MS3, we suggest that kittiwakes take advantage of the spatially and temporally predictable surface aggregations of sandeels at locations with higher CHLmax at maximum ebb (ME). In common with other studies, the kittiwakes are matching their foraging behaviour to that of their prey (e.g. Regular et al. 2010). The importance of chlorophyll concentration is likely to be an indirect link via the sandeels zooplankton prey and has been shown to provide a strong indication of the abundance of sandeels within the water column (Greenstreet et al. 2006). The lack of a link between the predictable surface aggregations of sandeels at ME and kittiwakes at MS2 is difficult to elucidate. At MS2, 100% of the foraging kittiwakes was found within MSFAs in contrast to MS1 and MS3 where 31–38% were foraging in areas without facilitation. MS2 also had higher overall chlorophyll and sandeel abundance but had significantly lower stratification, CHLmax biomass and tidal stratification log10(h/U3), indicating higher tidal mixing than either MS1 or MS3, and thus less likelihood of predictable surface prey aggregations. This suggests that subtle differences in vertical mixing that reduces stratification, such as small changes in depth and/or weather conditions, can cause dramatic differences in the availability of prey to specialised predators such as kittiwakes. These results also suggest that vertical water column characteristics can influence the presence and schooling behaviour of fish.

Functional principal component analysis enabled us to analyse repeated surveys conducted over the tidal/diurnal cycle to determine changes in bio-physical characteristics, sandeel schooling behaviour and kittiwake foraging behaviour with diurnal and tidal cycles. This is the first application of FDA to address issues associated with autocorrelated data collected with repeat-design surveys. FDA has some advantages over more correlative analysis techniques such as GLMMs or GAMMs by providing a visual as well as a statistical means of identifying spatio-temporal patterns. It allowed us to identify how spatial patterns changed with the tidal cycle and hence identify the times and locations of predator–prey interaction. We were thus able to identify possible mechanistic explanations for spatio-temporal patterns of predator–prey behaviour in relation to tidal current speed, subsurface chlorophyll concentrations and temperature stratification. It is possible that the conditions during this survey represent an extreme, it having been a particularly hot summer with strong stratification and this aspect of the survey having been carried out over 3 days. However, even from this limited study, it is clear that kittiwakes optimise their foraging to the conditions, preferentially foraging on predictable tidally influenced surface aggregations of sandeels, or foraging within MSFAs, providing the first indication of the mechanisms linking oceanography with kittiwake foraging. This is also the first application of this methodology to this type of data and the only study to address the autocorrelated nature of data collected with a repeated marine survey design.

Synthesis and Applications

The modelling framework presented here is appropriate for any data collected repeatedly from the same location over time. FDA provides a set of tools to analyse spatio-temporal patterns not only at the fine scale analysed here (i.e. 1–10 s of km and hours) but also at much larger scales (e.g. over continents and thousands of years: Cheddadi & Bar-Hen 2009). Combining FDA with marine data collected continuously from the same location surveyed repeatedly over diurnal and tidal cycles has proven highly effective for indentifying fine-scale mechanisms and has wide-ranging applicability for coupling marine bio-physical variables with predator–prey behaviour. The combined techniques provide a powerful tool for understanding fine-scale spatio-temporal dynamics of overdispersed mobile species in heterogeneous environments. Using these techniques to incorporate temporal factors into ecosystem research would improve our understanding of dynamic natural systems and provide a valuable contribution to conservation management. Indeed, understanding the mechanisms of bio-physical coupling has been identified as one of the key knowledge gaps within ecosystem-based approaches to fisheries management. (Botsford, Castilla & Peterson 1997; Barange 2005). Specifically, fisheries management relies on accurate stock assessments, which require a good understanding of the spatio-temporal dynamics of fish behaviour. However, fisheries surveys are often constrained by extreme variation in catch rates and this can lead to high uncertainty in population estimates. Here, we have identified potential sources of bias in sandeel stock assessment associated with varying tidal currents and fish behavioural responses. Subsequently, our findings indicate that previous surveys are likely to have underestimated sandeel abundance, because changes in fish aggregations in relation to tidal currents were not considered. Therefore, we recommend that future fisheries stock assessments incorporate temporal tidal factors and related fish behaviour into survey design.

Furthermore, we suggest that surveys designed to examine the use of habitat by animals in dynamic environments also consider temporal factors to accurately assess and understand the nature of bio-physical coupling. Importantly, locations of critical marine habitat, vital for identifying vulnerable populations’ conservation requirements, may be ephemeral in nature. In this case, we identified critical habitat at fine scales which only existed during very limited windows of the tidal cycle beyond the resolution of most management considerations. Temporal factors are also important at larger scales. Identifying precisely where and when vulnerable populations are particularly exposed to threats such as fisheries over-exploitation, climate change or interactions with marine renewable devices depends on understanding the temporal factors affecting behaviour, ranging patterns and habitat use. Finally, temporal considerations are vital for the effective identification, designation and management of MPAs and for the conservation management of predator and prey populations in all dynamic ecosystems where decisions based on short-term survey effort will have long-lasting implications.


This analysis was funded by the NERC ‘Sustainable Marine Bioresources’ funding. Funding for original fieldwork was via EU Q5RS 2000-30864; IMPRESS: Interactions between the Marine Environment, Predators and prey: implications for Sustainable Sandeel fisheries. Special thanks Crew of the Pelagia and especially Martin Laan and Santiago Gonsalez. Birder/sea mammal observers: Suzan van Lieshout, Luc Meeuwisse, Phillip Schwemmer, Nicole Sonntag; Volunteers for oceanography: Jackie Smith, Damion Nixon; and Dr. Patrick Holligan, University of Southampton for the analysis of the chlorophyll samples. Very special thanks to great efforts from John Dunn at FRS Marine Lab, Aberdeen (now Marine Scotland Science). We  would  also  like  to acknowledge the invaluable efforts of Oliver Ross and advice from Matthew Palmer. Simon Ingram, Lisa Ballance and several anonymous reviewers made useful comments on earlier versions of this script.